## 4. AMEs in the LPM for Transitions from Agriculture to Other Low-Skilled Industries

## Source details

**Canonical URL:** [4. AMEs in the LPM for Transitions from Agriculture to Other Low-Skilled Industries](https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021220-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2021/english/wpiea2021220-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2021/english/wpiea2021220-print-pdf.pdf.json)

---

### Introduction
- Research question: what is driving the shrinkage of agricultural employment in the Philippines — wage differentials or other factors?
- Key empirical observation: the agricultural exodus is a flow process from agriculture to non-agriculture with an intermediate step through non-employment (unemployed or non-participating).
- Main inference: shrinking agricultural employment can be largely attributed to increasing job separation in agriculture and greater job finding in non-agriculture; other economic and institutional factors also play important roles.

### Data and measurement
- Dataset: Philippines Labor Force Survey (LFS) Microdata Public Use Files.
- Sample coverage: nearly 11.4 million quarter-individual observations spanning 2005Q2 to 2019Q4, covering 17 administrative regions.
- Sample restrictions and notes:
  - Sample restricted to individuals age 15 to 64.
  - Employed workers excluding public administration, defense, compulsory social security, and extra-territorial organizations and bodies.
  - Unemployed are grouped with non-participants because the LFS does not distinguish unemployment from non-participation in the past-quarter employment question used for transitions.
  - Identification limitation: large N but small T; individuals cannot be uniquely identified across quarters due to lack of individual and geographic ID variables.
- Variables and construction:
  - EN rate (Employment to Non-employment): dependent variable = 1 when EN transition occurs, 0 when worker remains employed.
  - NE rate (Non-employment to Employment): dependent variable = 1 if employed in survey quarter, 0 otherwise.
  - Agricultural vs non-agricultural rates classified based on whether worker was employed in agriculture before non-employment (after non-employment).
  - Weighted averages use final person weights provided in each survey quarter.
  - Trends extracted using the Hodrick-Prescott Filter with smoothing parameter of 1600.

### Time series trends of EN and NE rates (facts)
- EN (job separation) rates:
  - EN rate is on average higher for agricultural workers than for non-agricultural workers.
  - EN rate for agricultural workers rises sharply in the latter half of the sample, reaching 7.3% at the end of the sample.
  - EN rate for non-agricultural workers remained relatively stable over the sample.
- NE (job finding) rates:
  - NE rate is on average lower in agriculture than in non-agriculture.
  - Both NE rates remain low before 2016 and then climb; the rise in NE for non-agriculture is significantly larger than for agriculture.
- Job-to-job transitions (EE) are limited:
  - Agricultural-to-non-agricultural EE transitions averaged 2.4% during the sample period.
  - Non-agricultural-to-agricultural EE transitions averaged 0.9% during the sample period.
- Complementary facts:
  - Agricultural employment share fell from 35% in 2005 to 24% in 2019.
  - On average 3.8% of employed workers each quarter are age 60 to 64.

### Theoretical model and identification of 'push' and 'pull' factors
- Model: two-market random search model (Mortensen–Pissarides framework) with agriculture and non-agriculture, capturing job finding, separation, and industry switching under frictions.
- Partial-equilibrium characterization for agriculture:
  - Reservation productivity (x_Agri) and labor market tightness (θ_Agri) determine equilibrium via intersection of:
    - Job separation (JS) curve — positively sloped in (x_Agri, θ_Agri) space.
    - Job creation (JC) curve — negatively sloped in (x_Agri, θ_Agri) space.
- Push-factor mechanisms (increase incentives to leave agriculture):
  - Higher non-agricultural productivity (wider wage differentials) raises agricultural outside options → higher EN and lower NE in agriculture.
  - Higher matching efficiency in non-agriculture or higher non-agricultural flow outside value similarly raise agricultural outside options → higher EN.
  - Improved matching efficiency in agriculture can still increase EN and lower NE if JC shift is smaller than JS shift.
- Pull-factor mechanisms (increase incentives to stay in agriculture):
  - Lower agricultural flow outside value (e.g., due to low education) lowers outside options → lower EN and higher NE in agriculture.
  - Weaker bargaining power of agricultural workers can lower separations and raise job finding if JC shifts more than JS.

### Empirically available factors analyzed
- Real wage differentials:
  - Agricultural TFP growth for the Philippines declined during 2001–2016 and turned negative from 2013 onward.
  - Real wage differentials constructed as the natural logarithm of differences in weighted average real hourly wages between non-agricultural and agricultural workers in each region.
  - Philippines’ economy average annual growth: 6.4% between 2010–2019 versus 4.6% between 2001–2009.
  - Agricultural workers’ basic pay is less than half that of a typical Filipino worker; agricultural workers represented roughly two of every three working poor in 2012.
- Labor market efficiency:
  - Measured using World Economic Forum Global Competitiveness Index components; constructed index is arithmetic average of five indices and shows an increasing trend.
- Road density (transport infrastructure):
  - National road density intensified over the last ten years per DPWH ATLAS 2018 and 2019.
  - National Capital Region (NCR) has 188.24 kilometers of roads per square kilometer of land area in 2019.
  - Reported road density panel statistic: 6.73 kilometers per square kilometer (and a regional low of 23.91 kilometer per square kilometer referenced elsewhere).
- Education and human capital:
  - High school and college education used as proxies; share with at least high school education: average of over 80% in non-agriculture and less than 50% in agriculture.
- Agricultural clusters:
  - Regional share of total palay and corn production used as proxy; correlation of 0.46 with agricultural share of real value-added for 2009-2018.

### Empirical findings (summary of regression evidence)
- Baseline LPM outcomes and dimensions:
  - Observations:
    - EN_Agri: 1,437,827
    - EN_Nonagri: 2,553,367
    - NE_Agri: 2,448,570
    - NE_Nonagri: 2,496,782
  - R-squared:
    - EN_Agri: 0.0536
    - EN_Nonagri: 0.0162
    - NE_Agri: 0.0383
    - NE_Nonagri: 0.0428
  - Joint Significance (F-stat):
    - EN_Agri: 333.57
    - EN_Nonagri: 178.40
    - NE_Agri: 173.48
    - NE_Nonagri: 144.22
- Selected Average Marginal Effects (AMEs) from baseline LPM (coefficient (standard error)):
  - Real Wage Differentials:
    - EN_Agri: 0.00947 ∗∗∗ (0.00245)
    - EN_Nonagri: -0.00572 ∗∗∗ (0.00203)
    - NE_Agri: -0.00683 ∗∗∗ (0.000991)
    - NE_Nonagri: 0.00631 ∗∗∗ (0.00191)
  - Labor Market Efficiency:
    - EN_Agri: 0.00675 ∗∗ (0.00279)
    - EN_Nonagri: -0.0175 ∗∗∗ (0.00289)
    - NE_Agri: -0.00174 (0.00167)
    - NE_Nonagri: 0.0496 ∗∗∗ (0.00191)
  - Road Density:
    - EN_Agri: 0.0487 ∗∗ (0.0213)
    - EN_Nonagri: -0.0170 (0.0192)
    - NE_Agri: -0.0768 ∗∗∗ (0.0115)
    - NE_Nonagri: 0.0552 ∗∗∗ (0.0180)
  - High School Education:
    - EN_Agri: 0.0204 ∗∗∗ (0.000473)
    - EN_Nonagri: 0.00276 ∗∗∗ (0.000469)
    - NE_Agri: -0.0190 ∗∗∗ (0.000342)
    - NE_Nonagri: -0.00237 ∗∗∗ (0.000475)
  - College Education:
    - EN_Agri: 0.0319 ∗∗∗ (0.000935)
    - EN_Nonagri: -0.0118 ∗∗∗ (0.000482)
    - NE_Agri: -0.0267 ∗∗∗ (0.000333)
    - NE_Nonagri: 0.00173 ∗∗∗ (0.000510)
  - Palay & Corn Production (regional share):
    - EN_Agri: -0.112 ∗∗∗ (0.0164)
    - EN_Nonagri: -0.0208 (0.0170)
    - NE_Agri: 0.00976 (0.00807)
    - NE_Nonagri: -0.0340 ∗∗ (0.0154)
- Key empirical interpretations:
  - Wage differentials (push) statistically explain variations in separations and findings consistent with the model.
  - Labor market efficiency and road density amplify wage-differential effects by lowering frictions and enabling switches across industries.
  - Lack of education (pull) limits agricultural workers’ ability to transition, reducing EN and increasing NE back into agriculture.
  - Regions with stronger agricultural clusters (higher palay and corn share) exhibit lock-in effects that reduce separations out of agriculture.

### Cross-region wage differentials and migration (selected results)
- Observations in cross-region specification:
  - EN_Agri: 1,437,160
  - NE_Agri: 2,380,033
  - EN_Nonagri: 2,551,595
  - NE_Nonagri: 2,491,172
- Selected AMEs (coefficient (standard error)):
  - Within-Region Real Wage Differentials:
    - EN_Agri: 0.0112 ∗∗∗ (0.00305)
    - NE_Agri: -0.0147 ∗∗∗ (0.00132)
    - EN_Nonagri: -0.00334 (0.00247)
    - NE_Nonagri: 0.0108 ∗∗∗ (0.00241)
  - Cross-Region Real Wage Differentials:
    - EN_Agri: 0.0106 ∗ (0.00546)
    - NE_Agri: 0.0328 ∗∗∗ (0.00222)
    - EN_Nonagri: -0.00945 ∗∗∗ (0.00362)
    - NE_Nonagri: 0.0232 ∗∗∗ (0.00348)
- Interpretation: cross-region wage gaps are important determinants, consistent with migration facilitating industry switching.

### EE transitions (job-to-job) — selected AMEs
- Observations:
  - EE_Agri→Nonagri: 138,668
  - EE_Nonagri→Agri: 224,502
- Selected AMEs (coefficient (standard error)):
  - Real Wage Differentials:
    - EE_Agri→Nonagri: -0.00168 (0.00183)
    - EE_Nonagri→Agri: -0.00298 ∗∗∗ (0.000750)
  - Labor Market Efficiency:
    - EE_Agri→Nonagri: 0.0121 ∗∗∗ (0.00164)
    - EE_Nonagri→Agri: -0.00872 ∗∗∗ (0.00184)
  - High School Education:
    - EE_Agri→Nonagri: 0.00749 ∗∗∗ (0.000340)
    - EE_Nonagri→Agri: -0.0103 ∗∗∗ (0.000238)
  - College Education:
    - EE_Agri→Nonagri: 0.00821 ∗∗∗ (0.000629)
    - EE_Nonagri→Agri: -0.0157 ∗∗∗ (0.000225)
  - Palay & Corn Production:
    - EE_Agri→Nonagri: -0.0314 ∗∗ (0.0129)
    - EE_Nonagri→Agri: -0.0283 ∗∗∗ (0.00630)
- Interpretation: labor market efficiency and education raise job-to-job transitions out of agriculture; wage differentials reduce job-to-job transitions into agriculture; agricultural clusters reduce EE mobility.

### Transitions from agriculture to low-skilled non-agricultural industries (selected AMEs)
- Definition: low-skilled industries defined as average share with at least high school education below 70%: mining and quarrying, construction, domestic and household services.
- Observations:
  - EN_Agri: 1,437,666
  - EN_Lowskilled: 394,678
  - NE_Agri: 2,430,806
  - NE_Lowskilled: 2,407,870
- Selected AMEs (coefficient (standard error)):
  - Real Wage Differentials:
    - EN_Agri: -0.00231 (0.00196)
    - EN_Lowskilled: -0.00515 (0.00473)
    - NE_Agri: -0.00334 ∗∗∗ (0.000750)
    - NE_Lowskilled: -0.00272 ∗∗∗ (0.000611)
  - Labor Market Efficiency:
    - EN_Agri: 0.0115 ∗∗∗ (0.00302)
    - EN_Lowskilled: -0.0351 ∗∗ (0.0165)
    - NE_Agri: -0.00193 (0.00161)
    - NE_Lowskilled: 0.0104 ∗∗∗ (0.000887)
  - Road Density:
    - EN_Agri: -0.00373 (0.0222)
    - EN_Lowskilled: 0.146 ∗∗ (0.0567)
    - NE_Agri: -0.0240 ∗∗ (0.0106)
    - NE_Lowskilled: 0.00528 (0.00788)
  - High School Education:
    - EN_Agri: 0.0204 ∗∗∗ (0.000473)
    - EN_Lowskilled: 0.00357 ∗∗∗ (0.00113)
    - NE_Agri: -0.0191 ∗∗∗ (0.000345)
    - NE_Lowskilled: -0.00534 ∗∗∗ (0.000277)
  - College Education:
    - EN_Agri: 0.0320 ∗∗∗ (0.000935)
    - EN_Lowskilled: 0.00621 ∗∗∗ (0.00167)
    - NE_Agri: -0.0269 ∗∗∗ (0.000335)
    - NE_Lowskilled: -0.0100 ∗∗∗ (0.000269)
  - Palay & Corn Production:
    - EN_Agri: -0.106 ∗∗∗ (0.0162)
    - EN_Lowskilled: -0.123 ∗∗ (0.0484)
    - NE_Agri: 0.0190 ∗∗ (0.00809)
    - NE_Lowskilled: 0.00159 (0.00727)
- Interpretation: labor market efficiency drives reallocation into low-skilled non-agricultural industries; education increases separations from agriculture but higher education is associated with lower job finding in low-skilled industries.

### Robustness checks and sensitivity analyses
- Time period interactions: adding time dummies for 2005Q2-2008Q1 and 2017Q1-2019Q4 and interacting with push/pull factors — results robust with no significant coefficient changes overall.
- Clustered standard errors at regional level: real wage differentials and road density lose some significance, but high school and college education remain explanatory.
- Logit specification: broadly similar results to the LPM.
- Economic significance exercise:
  - Finding: real wage differentials, while statistically significant, have limited economic impact relative to education and road density.
  - Road density and education (high school and college) are economically important — especially for raising EN_Agri and affecting NE outcomes.

### Policy recommendations and implications
- Maintain efficient labor markets to reduce frictions and support intersectoral reallocations.
- Invest in education:
  - Expanding high school coverage — shown sufficient to speed up reallocations from agriculture to non-agriculture.
  - Promoting education and upgrading skills of agricultural workers, especially young and female workers in regions with low education attainment.
- Expand transport infrastructure to lower reallocation costs and facilitate job search and matching across sectors.
- Continue land reform measures as a complement to labor-market and infrastructure policies (land reform effects not directly captured in this empirical study).
- Design policies to ensure preparedness of those who choose to leave agriculture to reduce scarring from shocks such as COVID-19.
- Consider geographic mobility costs and policies that reduce migration frictions, since cross-region real wage differentials are important determinants of separations/findings.

### Conclusion summary
- The agricultural exodus in the Philippines over the last 15 years is driven by increased separations from agriculture and redirected search effort to non-agriculture.
- Statistically significant push factors: widening wage differentials, increasing labor market efficiency, expanding road density.
- Statistically significant pull constraints: low high school and college attainment and regional agricultural clusters that dampen outflows.
- Economically, education (even high school) and road density matter more than wage differentials in facilitating labor reallocations from agriculture to non-agriculture.
- Empirical framework limited by data availability; extensions could include land reform progress and climate change effects when data permit.

*Source: Chapter 4 and appendices (pages 31– ) of wpiea2021220-print-pdf.*

### References .............................................................................................................

### References

### Figures
- 1. Agriculture Performance in Selected ASEAN Countries ...................................................4
- 2. Labor Market Composition .................................................................................................9
- 3. Time Series Trends of the Labor Market Transition Rates ..............................................10
- 4. Job Separation and Creation in the Agricultural Labor Market .......................................10
- 5. The Influence of Higher Non-Agricultural Productivity on Job Separation 
  and Finding ................................................................................................................13
- 6. The Influence of Lower Agricutural Productivity on Job Separation and Finding ..........14
- 7. The Influence of Better Matching in Agriculture on Job Separation and Finding ...........15
- 8. The Influence of Lower Flow Outside Value for Agricultural Workers on Job  
  Separation and Finding ..............................................................................................16
- 9. The Influence of Weaker Bargaining of Agricutural Workers on Job Separation 
  and Finding ................................................................................................................17
- 10. Real Wage Differentials ...................................................................................................18
- 11. Labor Market Efficiency Index ........................................................................................19
- 12. National Road Density ......................................................................................................20
- 13. Share of Employed Workers with at Least High School Education .................................21
- 14. Regional Share of Total Palay and Corn Production ........................................................22
- 15. Within-Region and Cross-Region Real Wage Differentials .............................................26
- 16. Time Series Trends of the EE Rates .................................................................................28
- 17. Average Share of Employed Workers with at Least High School Education,  
  by Industry .................................................................................................................30
- 18. Economic Significance .....................................................................................................33
- 19. AMEs of High School Education, by Region ...................................................................34
- 20. AMEs of High School Education, by Age ........................................................................34
- 21. AMEs of High School Education, by Gender ..................................................................35
- 22. AMEs of Road Density, by Region ..................................................................................35

### Tables
- 1. AMEs in the Baseline LPM ..............................................................................................24
- 2. AMEs in the LPM with Cross-Region Real Wage Differentials ......................................27
- 3. AMEs in the LPM for EE Transitions ..............................................................................29

*https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021220-print-pdf.pdf*

### 4. AMEs in the LPM for Transitions from Agriculture to Other Low-Skilled Industries ....31

### 4. AMEs in the LPM for Transitions from Agriculture to Other Low-Skilled Industries

### Introduction
- Research question: what is driving the shrinkage of agricultural employment in the Philippines — wage differentials or other factors?
- Key empirical observation: the agricultural exodus is a flow process from agriculture to non-agriculture with an intermediate step through non-employment (unemployed or non-participating).
- Main inference: shrinking agricultural employment can be largely attributed to increasing job separation in agriculture and greater job finding in non-agriculture; other economic and institutional factors also play important roles.

### Data and measurement
- Dataset: Philippines Labor Force Survey (LFS) Microdata Public Use Files.
- Sample coverage: nearly 11.4 million quarter-individual observations spanning 2005Q2 to 2019Q4, covering 17 administrative regions.
- Sample restrictions and notes:
  - Sample restricted to individuals age 15 to 64.
  - Employed workers excluding public administration, defense, compulsory social security, and extra-territorial organizations and bodies.
  - Unemployed are grouped with non-participants because the LFS does not distinguish unemployment from non-participation in the past-quarter employment question used for transitions.
  - Identification limitation: large N but small T; individuals cannot be uniquely identified across quarters due to lack of individual and geographic ID variables.
- Variables and construction:
  - EN rate (Employment to Non-employment): calculated from matched records starting with employed in the past quarter; dependent variable = 1 when EN transition occurs, 0 when worker remains employed.
  - NE rate (Non-employment to Employment): sample = those non-employed in past quarter; dependent variable = 1 if employed in survey quarter, 0 otherwise.
  - Agricultural vs non-agricultural rates are computed by classifying samples based on whether the worker was employed in agriculture before non-employment (after non-employment).
  - Weighted averages use final person weights provided in each survey quarter.
  - Trends extracted using the Hodrick-Prescott Filter with smoothing parameter of 1600.

### Time series trends of EN and NE rates (facts)
- EN (job separation) rates:
  - In levels, EN rate is on average higher for agricultural workers than for non-agricultural workers.
  - EN rate for agricultural workers rises sharply in the latter half of the sample, reaching 7.3% at the end of the sample.
  - EN rate for non-agricultural workers remained relatively stable over the sample.
- NE (job finding) rates:
  - NE rate is on average lower in agriculture than in non-agriculture.
  - Both NE rates remain low before 2016 and then climb; the rise in NE for non-agriculture is significantly larger than for agriculture.
- Job-to-job transitions (EE) are limited:
  - Agricultural-to-non-agricultural EE transitions averaged 2.4% during the sample period.
  - Non-agricultural-to-agricultural EE transitions averaged 0.9% during the sample period.
- Complementary demographic and labor statistics:
  - Agricultural employment share fell from 35% in 2005 to 24% in 2019.
  - On average 3.8% of employed workers each quarter are age 60 to 64.

### Theoretical model and identification of 'push' and 'pull' factors
- Model: two-market random search model (Mortensen–Pissarides framework) with agriculture and non-agriculture, capturing job finding, separation, and industry switching under frictions.
- Partial-equilibrium characterization for agriculture:
  - Reservation productivity (x_Agri) and labor market tightness (θ_Agri) determine equilibrium via intersection of:
    - Job separation (JS) curve — positively sloped in (x_Agri, θ_Agri) space.
    - Job creation (JC) curve — negatively sloped in (x_Agri, θ_Agri) space.
- Push-factor mechanisms (increase incentives to leave agriculture):
  - Higher non-agricultural productivity (wider wage differentials) shifts JS left → higher reservation productivity, lower θ → higher EN and lower NE in agriculture.
  - Higher matching efficiency in non-agriculture or higher non-agricultural flow outside value similarly raise agricultural outside options → higher EN.
  - Improved matching efficiency in agriculture can shift JS left and JC right; if JC shift is smaller, equilibrium still features higher EN and lower NE.
- Pull-factor mechanisms (increase incentives to stay in agriculture):
  - Lower agricultural flow outside value (e.g., due to low education) shifts JS right → lower reservation productivity, higher θ → lower EN and higher NE in agriculture.
  - Weaker bargaining power of agricultural workers (e.g., due to agricultural clusters or obsolete skills) can shift both JS and JC right; if JC shift is relatively small, equilibrium yields lower EN and higher NE.

### Empirically available factors analyzed
- Real wage differentials:
  - Agricultural TFP growth for the Philippines declined during 2001–2016 and turned negative from 2013 onward.
  - Real wage differentials are constructed as the natural logarithm of the differences in weighted average real hourly wages between non-agricultural and agricultural workers in each region.
  - The Philippines’ economy average annual growth: 6.4% between 2010–2019 versus 4.6% between 2001–2009 (context for non-agricultural gains).
  - Agricultural workers’ basic pay is less than half that of a typical Filipino worker; agricultural workers represented roughly two of every three working poor in 2012.
- Labor market efficiency:
  - Measured using the World Economic Forum Global Competitiveness Index components: flexibility of hiring and firing practices; cooperation in labor-employer relations; flexibility of wage determination; reliance on professional management; extent to which pay reflects employee productivity.
  - The constructed labor market efficiency index (arithmetic average of the five indices) shows an increasing trend during the sample period, indicating improving employment absorption in non-agriculture.
- Road density (transport infrastructure):
  - National road density intensified over the last ten years per DPWH ATLAS 2018 and 2019.
  - Wide regional disparity exists; National Capital Region (NCR) has 188.24 kilometers of roads per square kilometer of land area in 2019.
- Education and human capital:
  - High school and college education are used as proxies for human capital development and the ability of agricultural workers to access non-agricultural jobs.
- Agricultural clusters:
  - Each region’s share in total palay and corn production is used as a proxy for agricultural clusters.

### Empirical findings (summary)
- Regression analysis linking EN and NE rates to proposed factors shows:
  - Wage differentials (push) statistically help explain variations in job separation and finding rates, consistent with model predictions.
  - Labor market efficiency and transport infrastructure improvements amplify the effect of wage differentials by lowering frictions and enabling switches across industries.
  - Lack of education (pull) limits agricultural workers’ ability to transition to non-agricultural jobs, reducing EN and increasing NE back into agriculture.
  - Regions with stronger agricultural clusters (higher share in palay and corn production) make switching out of agriculture more difficult, reinforcing pull effects.
- Relative importance and economic significance:
  - Although increasing real wage differentials are statistically significant, they do not contribute much to the agricultural exodus in the Philippines relative to factors that facilitate switching industries.
  - High school education significantly raises agricultural workers’ job separation margin (comparatively more than effect on low-skilled non-agricultural industries).
  - College education further amplifies employment outflows by improving non-agricultural job finding and retention.
  - Road density is economically significant: higher road density substantially reduces job finding in agriculture and boosts job finding in non-agriculture.

### Policy recommendations and implications
- Maintain efficient labor markets to reduce frictions and support intersectoral reallocations.
- Invest in education, with emphasis on:
  - Expanding high school coverage — shown sufficient to speed up reallocations from agriculture to non-agriculture.
  - Promoting education and upgrading skills of agricultural workers, especially young and female workers in regions with low education attainment.
- Expand transport infrastructure to lower reallocation costs and facilitate job search and matching across sectors.
- Continue land reform measures as a complement to labor-market and infrastructure policies, acknowledging land reform effects are not directly captured in this empirical study.
- Design policies to ensure preparedness of those who choose to leave agriculture to reduce scarring from shocks such as COVID-19.

*Source: Chapter 4 (pages 31– ) of the provided IMF working paper PDF (wpiea2021220-print-pdf).*

### 23.91 kilometer per square kilometer. Cagayan Valley has the lowest road density among all, at

### wpiea2021220-print-pdf - 23.91 kilometer per square kilometer. Cagayan Valley has the lowest road density among all, at

### Road density

- 23.91 kilometer per square kilometer.
- Cagayan Valley has the lowest road density among all, at

*https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021220-print-pdf.pdf*

### 6.73 kilometers per square kilometer.

### wpiea2021220-print-pdf - 6.73 kilometers per square kilometer.

### Key statistics and descriptive facts
- Road density reported: 6.73 kilometers per square kilometer.
- National road density series panel labels include values around 9.6, 9.8, 10, 10.2, 10.4, 10.6 (figure axis values).
- Share of employed workers with at least high school education: average of over 80% in non-agriculture and less than 50% in agriculture (textual summary referring to Figure 13).
- Regional share of total palay and corn production used as a proxy; correlation of 0.46 with agricultural share of real value-added for 2009-2018 (proxy justification).
- Hodrick-Prescott Filter smoothing parameter used for trends: 1600.

### Empirical setup and methods
- Data sources:
  - Labor Force Survey, Philippine Statistics Authority.
  - 2018 & 2019 ATLAS, Department of Public Works and Highways (road density 2009-2019; BARMM imputed).
  - Agriculture, Forestry, Fisheries Database, Philippine Statistics Authority; Palay and Corn Production Survey (PCPS).
- Model: Linear Probability Model (LPM) for binary transition outcomes with heteroskedasticity-robust standard errors.
- Dependent outcomes studied: EN_Agri, EN_Nonagri, NE_Agri, NE_Nonagri, EE transitions (EE_Agri→Nonagri, EE_Nonagri→Agri).
- Covariates include: real wage differentials, labor market efficiency index, logged road density, high school education dummy, college education dummy, regional share of palay & corn production, demographic dummies (gender, age groups, marital status), region and year & quarter fixed effects.
- Imputation adjustment: BARMM assigned national average road density; 2005-2008 road density set equal to 2009 level; a dummy for imputed road density included (Cohen et al. (2013) approach referenced).
- Cross-region real wage differentials constructed for migration analysis (definitions differ for agricultural vs non-agricultural worker transition regressions).

### Baseline regression results (Table 1 — average marginal effects)
- Observations:
  - EN_Agri: 1,437,827
  - EN_Nonagri: 2,553,367
  - NE_Agri: 2,448,570
  - NE_Nonagri: 2,496,782
- R-squared:
  - EN_Agri: 0.0536
  - EN_Nonagri: 0.0162
  - NE_Agri: 0.0383
  - NE_Nonagri: 0.0428
- Joint Significance (F-stat):
  - EN_Agri: 333.57
  - EN_Nonagri: 178.40
  - NE_Agri: 173.48
  - NE_Nonagri: 144.22
- Selected AMEs (coefficient (standard error)):
  - Real Wage Differentials:
    - EN_Agri: 0.00947 ∗∗∗ (0.00245)
    - EN_Nonagri: -0.00572 ∗∗∗ (0.00203)
    - NE_Agri: -0.00683 ∗∗∗ (0.000991)
    - NE_Nonagri: 0.00631 ∗∗∗ (0.00191)
  - Labor Market Efficiency:
    - EN_Agri: 0.00675 ∗∗ (0.00279)
    - EN_Nonagri: -0.0175 ∗∗∗ (0.00289)
    - NE_Agri: -0.00174 (0.00167)
    - NE_Nonagri: 0.0496 ∗∗∗ (0.00191)
  - Road Density:
    - EN_Agri: 0.0487 ∗∗ (0.0213)
    - EN_Nonagri: -0.0170 (0.0192)
    - NE_Agri: -0.0768 ∗∗∗ (0.0115)
    - NE_Nonagri: 0.0552 ∗∗∗ (0.0180)
  - High School Education:
    - EN_Agri: 0.0204 ∗∗∗ (0.000473)
    - EN_Nonagri: 0.00276 ∗∗∗ (0.000469)
    - NE_Agri: -0.0190 ∗∗∗ (0.000342)
    - NE_Nonagri: -0.00237 ∗∗∗ (0.000475)
  - College Education:
    - EN_Agri: 0.0319 ∗∗∗ (0.000935)
    - EN_Nonagri: -0.0118 ∗∗∗ (0.000482)
    - NE_Agri: -0.0267 ∗∗∗ (0.000333)
    - NE_Nonagri: 0.00173 ∗∗∗ (0.000510)
  - Palay & Corn Production (regional share):
    - EN_Agri: -0.112 ∗∗∗ (0.0164)
    - EN_Nonagri: -0.0208 (0.0170)
    - NE_Agri: 0.00976 (0.00807)
    - NE_Nonagri: -0.0340 ∗∗ (0.0154)
- Demographic patterns (selected):
  - Female:
    - EN_Agri: 0.0925 ∗∗∗ (0.000627)
    - EN_Nonagri: 0.0115 ∗∗∗ (0.000343)
    - NE_Agri: -0.0311 ∗∗∗ (0.000276)
    - NE_Nonagri: -0.0424 ∗∗∗ (0.000445)
  - Age groups 25-34, 35-44, 45-54, 55-64 generally associated with lower separation probabilities relative to base age (negative AMEs on EN outcomes; positive AMEs on NE outcomes in many cases).
- Key interpretations from baseline:
  - Widening real wage differentials: positive and significant for EN_Agri and NE_Nonagri; negative and significant for EN_Nonagri and NE_Agri — interpreted as incentives to switch out of agriculture but also increases in non-employment flows.
  - Labor market efficiency and road density: positive and highly significant effects on separations out of agriculture and findings into non-agriculture.
  - High school and college education: both increase separations out of agriculture (EN_Agri) and reduce job finding into agriculture (NE_Agri); college education has a larger role than high school in driving outflows.
  - Regional palay & corn share: negative and significant effects on EN_Agri and NE_Nonagri — evidence of agricultural cluster lock-in.

### Cross-region wage differentials and migration (Table 2)
- Observations in cross-region specification:
  - EN_Agri: 1,437,160
  - NE_Agri: 2,380,033
  - EN_Nonagri: 2,551,595
  - NE_Nonagri: 2,491,172
- Selected AMEs:
  - Within-Region Real Wage Differentials:
    - EN_Agri: 0.0112 ∗∗∗ (0.00305)
    - NE_Agri: -0.0147 ∗∗∗ (0.00132)
    - EN_Nonagri: -0.00334 (0.00247)
    - NE_Nonagri: 0.0108 ∗∗∗ (0.00241)
  - Cross-Region Real Wage Differentials:
    - EN_Agri: 0.0106 ∗ (0.00546)
    - NE_Agri: 0.0328 ∗∗∗ (0.00222)
    - EN_Nonagri: -0.00945 ∗∗∗ (0.00362)
    - NE_Nonagri: 0.0232 ∗∗∗ (0.00348)
- Interpretation:
  - Cross-region real wage differentials are highly significant; larger cross-region gaps increase separations and findings in the agricultural sector and affect non-agricultural separations/findings, consistent with migration helping industry switching.

### EE transitions (job-to-job) (Table 3)
- Observations:
  - EE_Agri→Nonagri: 138,668
  - EE_Nonagri→Agri: 224,502
- Selected AMEs:
  - Real Wage Differentials:
    - EE_Agri→Nonagri: -0.00168 (0.00183)
    - EE_Nonagri→Agri: -0.00298 ∗∗∗ (0.000750)
  - Labor Market Efficiency:
    - EE_Agri→Nonagri: 0.0121 ∗∗∗ (0.00164)
    - EE_Nonagri→Agri: -0.00872 ∗∗∗ (0.00184)
  - High School Education:
    - EE_Agri→Nonagri: 0.00749 ∗∗∗ (0.000340)
    - EE_Nonagri→Agri: -0.0103 ∗∗∗ (0.000238)
  - College Education:
    - EE_Agri→Nonagri: 0.00821 ∗∗∗ (0.000629)
    - EE_Nonagri→Agri: -0.0157 ∗∗∗ (0.000225)
  - Palay & Corn Production:
    - EE_Agri→Nonagri: -0.0314 ∗∗ (0.0129)
    - EE_Nonagri→Agri: -0.0283 ∗∗∗ (0.00630)
- Interpretation:
  - Real wage differentials reduce job-to-job transitions into agriculture.
  - Labor market efficiency and education raise job-to-job transitions out of agriculture.
  - Agricultural cluster strength reduces EE mobility in both directions.

### Transitions from agriculture to low-skilled non-agricultural industries (Table 4)
- Low-skilled industries defined as average share with at least high school education below 70%: mining and quarrying, construction, domestic and household services.
- Observations:
  - EN_Agri: 1,437,666
  - EN_Lowskilled: 394,678
  - NE_Agri: 2,430,806
  - NE_Lowskilled: 2,407,870
- Selected AMEs:
  - Real Wage Differentials:
    - EN_Agri: -0.00231 (0.00196)
    - EN_Lowskilled: -0.00515 (0.00473)
    - NE_Agri: -0.00334 ∗∗∗ (0.000750)
    - NE_Lowskilled: -0.00272 ∗∗∗ (0.000611)
  - Labor Market Efficiency:
    - EN_Agri: 0.0115 ∗∗∗ (0.00302)
    - EN_Lowskilled: -0.0351 ∗∗ (0.0165)
    - NE_Agri: -0.00193 (0.00161)
    - NE_Lowskilled: 0.0104 ∗∗∗ (0.000887)
  - Road Density:
    - EN_Agri: -0.00373 (0.0222)
    - EN_Lowskilled: 0.146 ∗∗ (0.0567)
    - NE_Agri: -0.0240 ∗∗ (0.0106)
    - NE_Lowskilled: 0.00528 (0.00788)
  - High School Education:
    - EN_Agri: 0.0204 ∗∗∗ (0.000473)
    - EN_Lowskilled: 0.00357 ∗∗∗ (0.00113)
    - NE_Agri: -0.0191 ∗∗∗ (0.000345)
    - NE_Lowskilled: -0.00534 ∗∗∗ (0.000277)
  - College Education:
    - EN_Agri: 0.0320 ∗∗∗ (0.000935)
    - EN_Lowskilled: 0.00621 ∗∗∗ (0.00167)
    - NE_Agri: -0.0269 ∗∗∗ (0.000335)
    - NE_Lowskilled: -0.0100 ∗∗∗ (0.000269)
  - Palay & Corn Production:
    - EN_Agri: -0.106 ∗∗∗ (0.0162)
    - EN_Lowskilled: -0.123 ∗∗ (0.0484)
    - NE_Agri: 0.0190 ∗∗ (0.00809)
    - NE_Lowskilled: 0.00159 (0.00727)
- Interpretation:
  - Labor market efficiency is a key driver of reallocation into low-skilled non-agricultural industries.
  - Education increases separations from agriculture and reduces job finding in agriculture; however, higher education is associated with lower job finding in low-skilled industries (consistent with skilled workers switching away from low-skilled jobs).

### Robustness checks and sensitivity analyses
- Time period interactions: adding time dummies for 2005Q2-2008Q1 and 2017Q1-2019Q4 and interacting with push/pull factors — results robust with no significant coefficient changes overall.
- Clustered standard errors at regional level: real wage differentials and road density lose some significance, but high school and college education remain explanatory.
- Logit specification: broadly similar results to the LPM.
- Economic significance exercise: measured changes in EN/NE per standard deviation increase in continuous factors and AMEs for education dummies.
  - Finding: real wage differentials, while statistically significant, have limited economic impact relative to education and road density.
  - Road density and education (high school and college) are economically important — especially for raising EN_Agri (separations) and affecting NE outcomes.

### Policy-relevant findings and implications
- Education:
  - High school education significantly raises separation from agriculture (EN_Agri) and reduces job finding in agriculture (NE_Agri).
  - College education has larger effects than high school in driving outflows and increasing ability to find/keep non-agricultural jobs.
  - Less than 10% of employed agricultural workers reach college education, constraining outflows.
  - Policy implication: public investment in post-primary education, targeted to regions with low access to high school, can facilitate intersectoral reallocations.
- Transport infrastructure:
  - Road density increases separations from agriculture and boosts job finding in non-agriculture (NE_Nonagri); effects larger in regions with low initial road density.
  - Policy implication: transport infrastructure improvements in underdeveloped rural regions can promote reallocations and welfare gains.
- Labor market efficiency:
  - Improvements raise separations from agriculture and increase findings into non-agriculture.
  - Policy implication: reforms that improve labor market matching and reduce frictions help industry switching.
- Agricultural clusters:
  - Regions with higher shares of palay & corn production exhibit lock-in effects that reduce separations out of agriculture and reduce mobility.
  - Policy implication: targeted measures in strong agricultural clusters may be required to alleviate lock-in and broaden non-agricultural opportunities.
- Migration and cross-region wage gaps:
  - Cross-region real wage differentials are important determinants of separations/findings, indicating migration as a mechanism for industry switching.
  - Policy implication: consider geographic mobility costs and policies that reduce migration frictions when aiming to encourage intersectoral reallocations.
- Gender and demographic considerations:
  - Females separate more often and are less likely to find jobs in both sectors; education yields proportionally larger benefits for females (greater EN_Agri and NE_Nonagri effects with high school).
  - Older workers are less likely to separate (declining age profile of separations).

### Conclusion summary (paper conclusions)
- The agricultural exodus in the Philippines over the last 15 years is driven by increased separations from agriculture and redirected search effort to non-agriculture.
- Statistically significant push factors: widening wage differentials, increasing labor market efficiency, expanding road density.
- Statistically significant pull constraints: low high school and college attainment and regional agricultural clusters that dampen outflows.
- Economically, education (even high school) and road density matter more than wage differentials in facilitating labor reallocations from agriculture to non-agriculture.
- Empirical framework limited by data availability; extensions could include land reform progress and climate change effects when data permit.

*Source: wpiea2021220-print-pdf - 6.73 kilometers per square kilometer.*

### APPENDIX I.THEORETICAL MODEL

### APPENDIX I. THEORETICAL MODEL

### A. Environment — model structure and primitives
- Economy: unit mass of risk-neutral workers; infinite mass of identical, infinitely lived firms.
- Worker endowment: one unit of labor; each firm hires at most one worker.
- Production: y_j(x_j) = x_j, j ∈ {A, NA}, where x_j is industry-specific idiosyncratic productivity drawn from G_j(x_j) with support (0, ∞).
- Two separate matching markets (one for each industry). Matching function for industry j:
  - m_j(n_j, v_j) = ̄m_j n_j^{α_j} v_j^{1−α_j}, where ̄m_j is matching efficiency, α_j is elasticity w.r.t. non-employment, n_j is measure of non-employed, v_j is measure of vacancies.
  - Labor market tightness: θ_j ≡ v_j / n_j.
  - Job meeting rate: f_j(θ_j) ≡ m_j / n_j = ̄m_j θ_j^{1−α_j}.
  - Vacancy meeting rate: q_j(θ_j) ≡ m_j / v_j = ̄m_j θ_j^{−α_j}.
  - Identity: f_j(θ_j) = q_j(θ_j) θ_j.
- Productivity switches for employed workers with probability λ; new draw from G_j(.). Matches may be endogenously terminated if productivity is too low.
- Exogenous industry switching for workers with probability 1−γ_j.
- Empirical note (LFS): job-to-job transitions are small — 2.4% for agricultural workers and 1% for non-agricultural workers.

### B. Value functions — workers and firms
- Employed worker value in industry j, W_j(x_j):
  - W_j(x_j) = w_j(x_j) + β[(1−λ) W_j(x_j) + λ E_{x′_j}{ max[W_j(x′_j), γ_j N_j + (1−γ_j) N_¬j] }]
  - w_j(x_j) is current-period wage; β is discount factor; x′_j next-period productivity; N_j value of j-type non-employed; N_¬j value of non-employed in other industry.
- Non-employed worker value in industry j, N_j:
  - N_j = b_j + β[ f_j(θ_j) E_{x′_j}{ max[W_j(x′_j), γ_j N_j + (1−γ_j) N_¬j] } + (1−f_j(θ_j))[ γ_j N_j + (1−γ_j) N_¬j ] ]
  - b_j is flow outside value; f_j(θ_j) is job meeting rate.
- Firm match (filled position) value J_j(x_j):
  - J_j(x_j) = x_j − w_j(x_j) + β[ (1−λ) J_j(x_j) + λ E_{x′_j}[ max(J_j(x′_j), V_j) ] ]
  - V_j is value of unfilled position.
- Unfilled position value V_j:
  - V_j = −c_j + β[ q_j(θ_j) E_{x′_j}[ max(J_j(x′_j), V_j) ] + (1−q_j(θ_j)) V_j ], where c_j is per-period vacancy posting cost.
- Free entry drives V_j = 0, yielding job creation condition:
  - c_j / (β q_j(θ_j)) = E_{x′_j}[ max(J_j(x′_j), 0) ]  (Equation (6)).

### C. Surplus sharing and separation decision
- Match surplus when continued in industry j:
  - S_j(x_j) = J_j(x_j) + W_j(x_j) − [ γ_j N_j + (1−γ_j) N_¬j ].
- Surplus split by generalized Nash bargaining: worker takes fraction η ∈ (0,1) of total surplus; firm takes 1−η.
  - η J_j(x_j) = (1−η) [ W_j(x_j) − (γ_j N_j + (1−γ_j) N_¬j) ].
- Continuation/separation decision determined by cutoff productivity x_j such that S_j(x_j) = 0: below cutoff sever, above continue.

### D. Partial equilibrium in the agricultural labor market
- Goal: characterize job separation curve (JS) and job creation curve (JC) in (x_A, θ_A) plane and their slopes.
- Aggregate expected surplus for j-type matches: ES_j(x′_j) ≡ ∫_{x_j}^{∞} S_j(x′_j) dG_j(x′_j).
- Evolution of surplus for A-type match (Equation (8)):
  - S_A(x_A) = x_A − [ γ_A b_A + (1−γ_A) b_NA ] + β[ (1−λ) S_A(x_A) + λ ES_A(x′_A) ] − β η[ γ_A f_A(θ_A) ES_A(x′_A) + (1−γ_A) f_NA(θ_NA) ES_NA(x′_NA) ] − β(1−γ_A)(1−γ_A−γ_NA)(N_A − N_NA).
- Difference in non-employment values:
  - N_A − N_NA = b_A − b_NA + β η( f_A(θ_A) ES_A(x′_A) − f_NA(θ_NA) ES_NA(x′_NA) ) / (1 + β(1−γ_A−γ_NA)).
- Job separation condition evaluated at cutoff x_A (Equation (9)):
  - 0 = γ_A b_A + (1−γ_A) b_NA + β(1−γ_A)(1−γ_A−γ_NA)/(1 + β(1−γ_A−γ_NA)) (b_A − b_NA) − β λ ES_A(x′_A) + β η[ γ_A f_A(θ_A) ES_A(x′_A) + (1−γ_A) f_NA(θ_NA) ES_NA(x′_NA) ] + β(1−γ_A)(1−γ_A−γ_NA) β η( f_A(θ_A) ES_A(x′_A) − f_NA(θ_NA) ES_NA(x′_NA) )/(1 + β(1−γ_A−γ_NA)) − x_A.
- Under log-normal x_A with mean ̄x_A and standard deviation σ, expected surplus ES_A(x′_A) can be written (Equation (10)):
  - ES_A(x′_A) = (1/(1−β(1−λ))) e^{ ̄x_A + 1/2 σ^2 } { 1 − Φ[ ( log(x_A) − ( ̄x_A + σ^2 ) ) / σ ] } − x_A/(1−β(1−λ)) (1 − G_A(x_A)).
- Proposition 1 (job separation curve slope):
  - Under assumptions, dx_A/dθ_A (x_A,0, θ_A,0) > 0: job separation curve JS is positively sloped in (x_A, θ_A) plane.
  - Key signs used: ∂F/∂θ_A > 0 (Equation (12)); ∂F/∂x_A < 0 (Equation (13)); hence dx_A/dθ_A = −(∂F/∂θ_A)/(∂F/∂x_A) > 0.
- Job creation condition in agriculture (rewriting Equation (6), Equation (14)):
  - 0 = (1−η) ES_A(x′_A) − c_A / (β q_A(θ_A)).
- Proposition 2 (job creation curve slope):
  - Under assumptions, dx_A/dθ_A (x_A,0, θ_A,0) < 0: job creation curve JC is negatively sloped in (x_A, θ_A) plane.
  - Key signs: ∂H/∂θ_A = − c_A / (β ̄m_A) α_A θ_A^{α_A−1} < 0; ∂H/∂x_A = − (1−η) (1−G_A(x_A)) / (1−β(1−λ)) < 0; hence dx_A/dθ_A = −(∂H/∂θ_A)/(∂H/∂x_A) < 0.
- Equilibrium: intersection (x_A,0, θ_A,0) where JS and JC intersect; shifts in either curve (due to agricultural or non-agricultural labor-market changes) change equilibrium x_A and θ_A and therefore separation and finding rates.

### E. Comparative statics — push and pull factors affecting agricultural employment
- Definitions of observable rates:
  - Separation rate in agriculture: λ G_A(x_A) (monotonically increasing in x_A).
  - Job finding rate in agriculture: f_A(θ_A) (1 − G_A(x_A)) (increasing in θ_A, decreasing in x_A).

- Push factors (incentives to leave agriculture)
  - Increase in non-agricultural mean productivity ̄x_NA (raises outside option f_NA(θ_NA) ES_NA(x′_NA)):
    - For given θ_A, reservation productivity x_A rises → JS curve shifts left.
    - New equilibrium: x_A increases to x_A,1; θ_A falls to θ_A,1.
    - Outcomes: higher job separation in agriculture (λ G_A(x_A) rises) and lower job finding (f_A(θ_A)(1−G_A(x_A)) falls).
    - Interpretation: widening productivity/wage differentials incentivize agricultural workers to search for non-agricultural jobs.
  - Similar effects arise from increases in b_NA or ̄m_NA (they raise outside option for agricultural workers).
  - Decrease in ̄x_A can also produce leftward shifts in both JS and JC and similar net outcomes (see text).
  - Increase in agricultural matching efficiency ̄m_A:
    - Triggers leftward shift of JS and rightward shift of JC (JC shifts right because q_A(θ_A) = ̄m_A θ_A^{−α_A}).
    - If JC shift is relatively small (common when α_A is small at low θ_A), new equilibrium has x_A increase to x_A,3 and θ_A fall to θ_A,3.
    - Outcomes: greater job separation and lower job finding in agriculture.
    - Policy interpretation: better functioning matching in agriculture (e.g., improved transport infrastructure improving ̄m_A) can nonetheless promote agricultural employment outflows.

- Pull factors (incentives to remain in agriculture)
  - Decrease in agricultural flow outside value b_A:
    - For given θ_A, reservation productivity x_A falls → JS curve shifts right to JS*.
    - New equilibrium: x_A decreases to x_A,4; θ_A increases to θ_A,4.
    - Outcomes: job separation rate falls and job finding rate rises in agriculture.
    - Interpretation: lack of education or skill depreciation that lowers b_A can trap workers in agriculture (less likely to separate).
  - Decline in worker bargaining power η:
    - Mechanism differs: leads to rightward shifts in both JS and JC (text shows multiplicative terms involving η).
    - If JC shift is relatively small (often when α_A is large at high θ_A), new equilibrium at intersection of JS** and JC** yields lower job separation and higher job finding probabilities.
    - Interpretation: weaker worker bargaining power increases firms’ share (1−η), promoting job creation and reducing separations; thus inadequate education and agricultural clusters can dampen agricultural exodus.

- Channel distinctions emphasized:
  - Different push factors operate through different channels (raising outside options, improving matching efficiency, lowering ̄x_A) and can produce similar qualitative outcomes (higher separations, lower finds) but via distinct shifts of JS and/or JC.
  - Pull factors can reduce outflows either by lowering outside options (b_A down) or by weakening bargaining power (η down), but the implied shifts of JS and JC differ and produce distinct comparative-static paths.

*APPENDIX I. THEORETICAL MODEL*

### APPENDIX II.END BEHAVIORS OF THEENANDN ERATES

### APPENDIX II.END BEHAVIORS OF THEENANDN ERATES

### Purpose and identification of end-behavior issue
- Objective: Explore the extent to which baseline regression results are affected by the nonconforming end behaviors of EN and N E rates documented in Figures 3a and 3b.
- Approach: Apply two exercises by adding a time dummy indicating 2005Q2-2008Q1 and 2017Q1-2019Q4 respectively in the Linear Probability Model (LPM) for each labor market transition rate, and interact it with proposed ‘push’ and ‘pull’ factors.

### Estimation setup
- Models: LPM including interactions between time dummies (2005Q2-2008Q1 or 2017Q1-2019Q4) and the proposed factors.
- Controls: Demographic, region, and year & quarter dummies are included.
- Standard errors: Heteroskedasticity robust (Tables II.1 and II.2); clustered at regional level in related robustness checks (Appendix III).

### Key findings — robustness across sample splits (Table II.1: time dummy 2005Q2-2008Q1)
- General: Most results are robust across the two sample periods 2005Q2-2008Q1 and 2008Q2-2019Q4.
- Real Wage Differentials:
  - 2005Q2-2008Q1: EN_Agri = 0.0168 ∗∗∗; EN_Nonagri = -0.00603; N E_Agri = -0.0224 ∗∗∗; N E_Nonagri = -0.0240 ∗∗∗ (standard errors: (0.00580), (0.00546), (0.00225), (0.00356)).
  - 2008Q2-2019Q4: EN_Agri = 0.0106 ∗∗∗; EN_Nonagri = -0.00682 ∗∗∗; N E_Agri = -0.00812 ∗∗∗; N E_Nonagri = 0.0171 ∗∗∗ (standard errors: (0.00308), (0.00241), (0.00128), (0.00245)).
- Labor Market Efficiency:
  - 2005Q2-2008Q1: EN_Agri = -0.0111; EN_Nonagri = -0.0133 ∗∗; N E_Agri = 0.00411; N E_Nonagri = 0.0193 ∗∗∗ (s.e.: (0.00763), (0.00625), (0.00345), (0.00427)).
  - 2008Q2-2019Q4: EN_Agri = 0.0407 ∗∗∗; EN_Nonagri = -0.0138 ∗∗; N E_Agri = 0.00356; N E_Nonagri = 0.0586 ∗∗∗ (s.e.: (0.00698), (0.00582), (0.00328), (0.00397)).
- Road Density:
  - 2005Q2-2008Q1: EN_Agri = 0.152 ∗∗∗; EN_Nonagri = 0.00601; N E_Agri = -0.108 ∗∗∗; N E_Nonagri = 0.0281 (s.e.: (0.0257), (0.0231), (0.0136), (0.0218)).
  - 2008Q2-2019Q4: EN_Agri = 0.151 ∗∗∗; EN_Nonagri = 0.0101; N E_Agri = -0.103 ∗∗∗; N E_Nonagri = 0.0240 (s.e.: (0.0253), (0.0231), (0.0135), (0.0215)).
- High School Education:
  - 2005Q2-2008Q1: EN_Agri = 0.0218 ∗∗∗; EN_Nonagri = 0.00248 ∗∗∗; N E_Agri = -0.00884 ∗∗∗; N E_Nonagri = 0.00364 ∗∗∗ (s.e.: (0.000953), (0.000911), (0.000480), (0.000639)).
  - 2008Q2-2019Q4: EN_Agri = 0.0201 ∗∗∗; EN_Nonagri = 0.00281 ∗∗∗; N E_Agri = -0.0216 ∗∗∗; N E_Nonagri = -0.00379 ∗∗∗ (s.e.: (0.000545), (0.000537), (0.000406), (0.000564)).
- College Education:
  - 2005Q2-2008Q1: EN_Agri = 0.0332 ∗∗∗; EN_Nonagri = -0.0110 ∗∗∗; N E_Agri = -0.0167 ∗∗∗; N E_Nonagri = 0.00271 ∗∗∗ (s.e.: (0.00179), (0.000906), (0.000470), (0.000699)).
  - 2008Q2-2019Q4: EN_Agri = 0.0316 ∗∗∗; EN_Nonagri = -0.0120 ∗∗∗; N E_Agri = -0.0293 ∗∗∗; N E_Nonagri = 0.00138 ∗∗ (s.e.: (0.00109), (0.000555), (0.000395), (0.000608)).
- Palay & Corn Production:
  - 2005Q2-2008Q1: EN_Agri = -0.172 ∗∗∗; EN_Nonagri = 0.04190.0212; N E_Agri = -0.125 ∗∗∗ (s.e.: (0.0359), (0.0372), (0.0151), (0.0255)) — note source formatting shows "0.04190.0212" for EN_Nonagri and N E_Agri entries.
  - 2008Q2-2019Q4: EN_Agri = -0.105 ∗∗∗; EN_Nonagri = -0.00510; N E_Agri = 0.00736; N E_Nonagri = -0.0788 ∗∗∗ (s.e.: (0.0205), (0.0208), (0.0104), (0.0199)).
- Observations: EN_Agri = 1437827; EN_Nonagri = 2553367; N E_Agri = 2448570; N E_Nonagri = 2496782.

### Key findings — alternative split near sample end (Table II.2: time dummy 2017Q1-2019Q4)
- General: Estimated effects of several factors are robust for 2005Q2-2016Q4, while notable deviations appear in 2017Q1-2019Q4 for some variables.
- Real Wage Differentials:
  - 2005Q2-2016Q4: EN_Agri = 0.0164 ∗∗∗; EN_Nonagri = -0.000583; N E_Agri = -0.00456 ∗∗∗; N E_Nonagri = -0.00452 ∗∗∗ (s.e.: (0.00250), (0.00224), (0.000818), (0.00160)).
  - 2017Q1-2019Q4: EN_Agri = 0.00609; EN_Nonagri = -0.00710; N E_Agri = 0.00458; N E_Nonagri = 0.0320 ∗∗∗ (s.e.: (0.00916), (0.00552), (0.00363), (0.00628)).
- Labor Market Efficiency:
  - 2005Q2-2016Q4: EN_Agri = 0.117 ∗; EN_Nonagri = -0.160 ∗∗∗; N E_Agri = -0.210 ∗∗∗; N E_Nonagri = -0.272 ∗∗∗ (s.e.: (0.0663), (0.0381), (0.0208), (0.0462)).
  - 2017Q1-2019Q4: EN_Agri = 0.162 ∗∗∗; EN_Nonagri = -0.119 ∗∗∗; N E_Agri = -0.192 ∗∗∗; N E_Nonagri = -0.191 ∗∗∗ (s.e.: (0.0628), (0.0345), (0.0193), (0.0428)).
- Road Density:
  - 2005Q2-2016Q4: EN_Agri = -0.172 ∗∗∗; EN_Nonagri = -0.0900 ∗∗∗; N E_Agri = 0.00491; N E_Nonagri = 0.0656 ∗∗∗ (s.e.: (0.0289), (0.0262), (0.0120), (0.0174)).
  - 2017Q1-2019Q4: EN_Agri = -3.583 ∗∗∗; EN_Nonagri = -1.252 ∗; N E_Agri = -0.0503; N E_Nonagri = -1.552 ∗∗ (s.e.: (1.077), (0.723), (0.565), (0.739)).
- High School Education:
  - 2005Q2-2016Q4: EN_Agri = 0.0199 ∗∗∗; EN_Nonagri = 0.00328 ∗∗∗; N E_Agri = -0.00761 ∗∗∗; N E_Nonagri = 0.00337 ∗∗∗ (s.e.: (0.000477), (0.000467), (0.000255), (0.000339)).
  - 2017Q1-2019Q4: EN_Agri = 0.0231 ∗∗∗; EN_Nonagri = 0.00100; N E_Agri = -0.0549 ∗∗∗; N E_Nonagri = -0.0202 ∗∗∗ (s.e.: (0.00156), (0.00131), (0.00114), (0.00159)).
- College Education:
  - 2005Q2-2016Q4: EN_Agri = 0.0341 ∗∗∗; EN_Nonagri = -0.0109 ∗∗∗; N E_Agri = -0.0140 ∗∗∗; N E_Nonagri = 0.00185 ∗∗∗ (s.e.: (0.000968), (0.000479), (0.000251), (0.000367)).
  - 2017Q1-2019Q4: EN_Agri = 0.0225 ∗∗∗; EN_Nonagri = -0.0151 ∗∗∗; N E_Agri = -0.0694 ∗∗∗; N E_Nonagri = 0.00163 (s.e.: (0.00278), (0.00135), (0.00114), (0.00179)).
- Palay & Corn Production:
  - 2005Q2-2016Q4: EN_Agri = -0.133 ∗∗∗; EN_Nonagri = -0.00649; N E_Agri = 0.00918; N E_Nonagri = -0.000699 (s.e.: (0.0167), (0.0163), (0.00651), (0.0114)).
  - 2017Q1-2019Q4: EN_Agri = -0.0402; EN_Nonagri = -0.0850; N E_Agri = -0.0514; N E_Nonagri = -0.339 ∗∗∗ (s.e.: (0.0636), (0.0587), (0.0395), (0.0722)).
- Observations: EN_Agri = 1437827; EN_Nonagri = 2553367; N E_Agri = 2448570; N E_Nonagri = 2496782.

### Interpretation and conclusion
- Main conclusion: Most estimated effects of the proposed ‘push’ and ‘pull’ factors are robust to sample splits; however, notable inconsistencies concentrate on the behavior of N E_Nonagri towards the end of the sample period.
- Possible cause: The inconsistencies for N E_Nonagri may result from the implementation of the New Master Sample Design of the Labor Force Survey (LFS) starting in 2016Q2.
- Specific observations:
  - Negative and highly significant AME of real wage differentials on N E_Nonagri during 2005Q2-2008Q1 (Table II.1) aligns with observed stagnation of real wage differentials in Figure 10.
  - In Table II.2, road density exhibits large negative coefficients in 2017Q1-2019Q4 (e.g., EN_Agri = -3.583 ∗∗∗), suggesting road density may be acting as a proxy for other unobserved factors in those split samples.

*Source: wpiea2021220-print-pdf - APPENDIX II.END BEHAVIORS OF THEENANDN ERATES (IMF Working Paper PDF).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021220-print-pdf.pdf_